DMC-VB: DeepMind Control Vision Benchmark, Locomotion and Ant Maze Episodes with Visual Distractors for Offline RL (Google DeepMind)

Provider
Google DeepMind (Joseph Ortiz, Antoine Dedieu et al.)
License
CC-BY-4.0 (source)
Stored at
Google Cloud Storage, gs://dmc_vision_benchmark/dmc_vision_benchmark (us)
Access
Free and public. No account with the data provider is needed.
Category
Robotics
Source
https://github.com/google-deepmind/dmc_vision_benchmark

About this dataset

DMC-VB is an offline dataset for representation learning for control from pixels in the presence of visual distractors, built on the DeepMind Control Suite (MuJoCo physics). Locomotion data (locomotion/cheetah_run, walker_walk, humanoid_walk) cover four behavioral policy levels (expert, medium, mixed, random) and three distractor settings (none, static, dynamic background videos), 36 subsets of 2,000 episodes (1M steps) each. Ant maze data (antmaze_fixed_hidden_goal, antmaze_random_hidden_goal, antmaze_random_visible_goal) cover 7 maze layouts (empty, three single-obstacle, three multi-obstacle) with expert and medium policies, randomized or fixed start and goal locations and four floor and wall textures, 1M steps per subset. Episodes include pixel observations, proprioceptive states, actions and rewards. The whole bucket, including Kubric MOVi-D background videos at kubric_movi-d/, is about 950 GB. Used for offline reinforcement learning, behavior cloning and visual pretraining for robot locomotion and navigation. Loaders are in google-deepmind/dmc_vision_benchmark. License: CC BY 4.0. Cite Ortiz et al., DMC-VB: A Benchmark for Representation Learning for Control with Visual Distractors, 2024.

How to access this dataset in Python

import gcsfs
fs = gcsfs.GCSFileSystem(token='anon')
fs.ls('dmc_vision_benchmark/dmc_vision_benchmark/', detail=False)[:10]

Lists the files with anonymous access. Runs on any machine, with no cloud account.

Data structure

Files

Example file: dmc_vision_benchmark/antmaze_fixed_hidden_goal/easy7x7a/

Analyze this dataset without downloading it

On Scigantic, DMC-VB: DeepMind Control Vision Benchmark, Locomotion and Ant Maze Episodes with Visual Distractors for Offline RL (Google DeepMind) opens in a hosted JupyterLab notebook with the data already mounted, so you can read files where they are stored instead of copying them first. Each notebook includes an AI assistant that can see the mounted files and write and run analysis code. Open the dataset page to launch a notebook, or browse the full catalog.

This page is also available as Markdown and as JSON.

Related Robotics datasets